Building a research-software catalog with a coding agent: from hackathon prototype to public deployment

The researchers used generative AI and a coding agent to develop a prototype of research software catalog during a three-day hackathon, and completed the engineering work required to make it suitable for public deployment. The team evaluated the prototype through adversarial review, data quality checks, and browser-level validation, then applied these experiences to the MateriApps retrieval agent under development. This agent combines curated portal metadata, external documents, vector search, and local language model generation to produce content. Although the implementation was rapid, achieving reliable operation required significant additional engineering efforts. The main challenges were incomplete data acquisition, misleading evaluations, and output errors caused by silent failures. Observations indicate that AI-assisted software portals require explicit verification, monitoring, and repeated review, and curated metadata and maintained documents remain crucial. The development of MateriApps is still in the exploratory stage and is actively being developed.